Machine learning-based residual drift prediction of concrete-filled steel tube columns under earthquake loads

Abstract

This study introduces a machine learning (ML) model for predicting the residual (permanent) drift in concrete-filled steel tube (CFT) structures exposed to excessive loads. The relationship between the residual drift and design parameters of the CFT column was established through the adaptation of six distinct ML models. The accuracy of the proposed model was evaluated by comparing its predictions with the results derived from existing numerical models and code provisions. The proposed ML model surpasses existing residual drift prediction models by at least 4% in terms of R-squared value. To evaluate the effectiveness of the model, a nonlinear dynamic analysis of a three-story, three-bay CFT moment-resisting frame subjected to 44 ground motions was conducted using a numerical model. Additionally, two modified frame models were considered to investigate the influence of frame configuration on the performance of the ML model. Subsequently, the residual inter-story drift ratio derived from the seismic analysis using the numerical model was compared with the results obtained from the ML predictive model and code provisions. The proposed ML model exhibited greater accuracy in predicting the residual inter-story drift ratio than the code provisions when applied to the results of numerical analyses performed for weak-column strong-beam frames. The ML model showed a positive R-squared value, whereas the code provisions yielded a lower R-squared value. In turn, the ML model applied to the strong-column weak-beam moment-resisting frames tended to overestimate the residual inter-story drift ratio, with a ratio of the predicted value to the value obtained from the analysis equal to 3.5.